Skip to main content
This agent is currently in preview and is not available to all customers at this time. Contact Sales for more information.
Learns how an asset or system behaves directly from unlabeled sensor data, segmenting its behavior into discrete states and estimating how it moves between them, with no labels or feature engineering required. Every timestep in a batch file comes back with a state label, along with a compact map of the states the system visits and the transitions between them.
This agent is currently only available in batch mode; serving mode is not yet supported.

Use Cases

  • Map operating states without labeled data: Segments raw sensor behavior into discrete states with no ground truth, so a new asset gets a behavior model without weeks of feature engineering and labeling.
  • Surface hidden or degraded modes of operation: Finds the states the data actually contains, not only the ones engineers expect, so unknown or degrading behavior shows up instead of blending into a hand-defined state.
  • Bootstrap state labels for monitoring: Returns the input files with a state column added and uses any partial labels to name states, so teams can review, name, and carry discovered states into ongoing monitoring.